CN102598127A - Signal processing method, information processor, and signal processing program - Google Patents

Signal processing method, information processor, and signal processing program Download PDF

Info

Publication number
CN102598127A
CN102598127A CN2010800504499A CN201080050449A CN102598127A CN 102598127 A CN102598127 A CN 102598127A CN 2010800504499 A CN2010800504499 A CN 2010800504499A CN 201080050449 A CN201080050449 A CN 201080050449A CN 102598127 A CN102598127 A CN 102598127A
Authority
CN
China
Prior art keywords
noise
signal
noise information
deterioration
information
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN2010800504499A
Other languages
Chinese (zh)
Other versions
CN102598127B (en
Inventor
杉山昭彦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NEC Corp
Original Assignee
NEC Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by NEC Corp filed Critical NEC Corp
Publication of CN102598127A publication Critical patent/CN102598127A/en
Application granted granted Critical
Publication of CN102598127B publication Critical patent/CN102598127B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
    • G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
    • G10L21/0208—Noise filtering

Landscapes

  • Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Quality & Reliability (AREA)
  • Signal Processing (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Noise Elimination (AREA)
  • Soundproofing, Sound Blocking, And Sound Damping (AREA)

Abstract

Provided is a noise suppressing technology capable of suppressing various noises including unknown noises without storing information relating to a large number of noises in advance. Noises in a degraded signal are suppressed, and noise information is generated on the basis of noise suppression results. The noises in the degraded signal are suppressed using the generated noise information.

Description

Signal processing method, signal conditioning package and be used for the storage medium of storage signal handling procedure
Technical field
The present invention relates to suppress noise in the deterioration signal to strengthen the signal processing technology of echo signal.
Background technology
Noise reduction techniques is considered to partly a kind of or suppresses the signal processing technology of the signal (through strengthening the signal that echo signal obtains) that noise and output in the deterioration signal (signal that mixes that comprises noise and echo signal) strengthens fully.For example, noise suppressor is the system that a kind of inhibition is blended in the noise in the target audio signal.Noise suppressor is used in the various voice frequency terminals such as mobile phone.
About this type of technology, patent documentation 1 discloses and has a kind ofly suppressed the method for noise through input signal being multiply by gain less than 1 spectrum.Patent documentation 2 discloses a kind of through directly from deterioration signal, deducting the method that estimated noise suppresses noise.
The Technology Need that patent documentation 1 and 2 is described is from because the noise that mixes and estimating noise the echo signal of deterioration.Yet, only limited from the degree of accuracy of deterioration signal estimating noise.Therefore, the method described of patent documentation 1 and 2 is only effective during much smaller than echo signal at noise.If do not satisfy the condition of noise much smaller than echo signal, then the Noise Estimation degree of accuracy is relatively poor.Therefore, the method that patent documentation 1 and 2 is described can not obtain sufficient noise suppression effect, and the signal that strengthens includes big distortion.
On the other hand, patent documentation 3 discloses a kind of like this noise suppressing system, even it also can realize the less distortion in sufficient noise suppression effect and the enhancing signal at noise when the condition of echo signal is not being met.Suppose to know in advance to a certain extent that then the method for patent documentation 3 descriptions deducts the noise information (about the information of noisiness) of precedence record from deterioration signal, thereby suppress noise the characteristic that is mixed to the noise in the echo signal.Patent documentation 3 also discloses following method: if bigger through the input signal power of analyzing the input signal acquisition; Then big coefficient is integrated in the noise information; If perhaps input signal power is less, the coefficient that then integration is little and deduct integral result from deterioration signal.
[reference listing]
[PTL 1] Jap.P. No.4282227
[PTL 1] Jap.P. discloses No.8-221092 in early days
[PTL 1] Jap.P. discloses No.2006-279185 in early days
Summary of the invention
Yet above-mentioned patent documentation 3 disclosed layouts need be stored noisiness information in advance, and it is extremely limited to eliminate the type of noise.In order to increase the type that can eliminate noise, many noise informations need be by record.This has increased the manufacturing cost of needed memory size and device.In addition, patent documentation 3 disclosed technology can't suppress the unknown noise different with the storage noise information.
The present invention considers said circumstances and by being made, and its illustrative purpose provide the signal processing technology that addresses the above problem.
In order to realize above-mentioned illustrative purpose, the signal processing method of an illustrative aspects comprises according to the present invention: during noise in suppressing deterioration signal, generate noise information according to the squelch result to said deterioration signal; And utilize the noise information that is generated to suppress the noise in the said deterioration signal.
In order to realize above-mentioned illustrative purpose, the signal conditioning package of another illustrative aspects comprises according to the present invention: noise suppressor, and this noise suppressor suppresses the noise in deterioration signal; And the noise information generation unit, this noise information generation unit is based on generating noise information to the Noise Suppression result in the said deterioration signal, and wherein, said noise suppressor utilizes said noise information to suppress the noise in the said deterioration signal.
In order to realize above-mentioned illustrative purpose; The signal stored handling procedure is carried out computing machine in the non-instantaneity medium of the computer-readable of another illustrative aspects according to the present invention: generate the processing of noise information based on the process result that suppresses noise, and utilize the noise information that is generated to suppress the processing of the noise in the deterioration signal.
The beneficial effect of the invention
According to the present invention, can be provided at the signal processing technology of the various noises that suppress to comprise unknown noise under the situation that need not to store in advance many noise informations.
Description of drawings
Fig. 1 is the block diagram of schematic arrangement that the Noise Suppression Device 100 of first exemplary embodiment according to the present invention is shown;
Fig. 2 is the block diagram that the layout of FFT (Fast Fourier Transform (FFT)) unit 2 in the Noise Suppression Device 100 that is included in first exemplary embodiment according to the present invention is shown;
Fig. 3 is the block diagram that the layout of IFFT (inverse fast fourier transform) unit 4 in the Noise Suppression Device 100 that is included in first exemplary embodiment according to the present invention is shown;
Fig. 4 is the block diagram of schematic arrangement that the Noise Suppression Device 200 of the 3rd exemplary embodiment according to the present invention is shown;
Fig. 5 is the block diagram of schematic arrangement that the Noise Suppression Device 300 of the 4th exemplary embodiment according to the present invention is shown;
Fig. 6 is the block diagram of schematic arrangement that the Noise Suppression Device 400 of the 5th exemplary embodiment according to the present invention is shown;
Fig. 7 is the schematic block diagram of computing machine 1000 of the execution signal handler of the another exemplary embodiment according to the present invention; And
Fig. 8 is the block diagram that illustrates according to the layout example of signal conditioning package 1200 of the present invention.
Embodiment
Pass through example detailed description exemplary embodiment referring now to accompanying drawing.Notice that the composed component of describing in the exemplary embodiment only is an example, technical scope is not limited by following exemplary embodiment.
(first exemplary embodiment)
< general layout >
As first exemplary embodiment that is used to realize signal processing method; Noise Suppression Device will be described, this device partly or fully suppresses the noise in the deterioration signal (signal that mixes that comprises noise and echo signal) and exports enhancing signal (through strengthening the signal that echo signal obtains).Fig. 1 is the block diagram that the general layout of Noise Suppression Device 100 is shown.Noise Suppression Device 100 is as the part of the equipment such as digital camera, notebook computer or mobile phone.Yet this exemplary embodiment is not limited thereto and is applicable to the signal conditioning package that need remove any kind of noise from input signal.Fig. 8 is the block diagram that illustrates according to the layout example of the signal conditioning package 1200 of this exemplary embodiment.Signal conditioning package 1200 comprises noise suppression unit 3 and noise information generation unit 7.
Deterioration signal (signal that has mixed echo signal and noise) is transfused to input terminal 1 as sample value sequence.2 pairs of FFT unit offer the deterioration signal of input terminal 1 and carry out the conversion such as Fourier transform, thereby division of signal is become a plurality of frequency contents.The amplitude spectrum that noise suppression unit 3 receives in a plurality of frequency contents, and IFFT unit 4 is provided phase spectrum.Notice that in this situation, amplitude spectrum is provided for noise suppression unit 3.Yet this exemplary embodiment is not limited thereto, and can be provided for noise suppression unit 3 with square corresponding power spectrum of amplitude spectrum.
Temporary storage 6 comprises memory component and the storage noise information (about the information of noisiness) such as semiconductor memory.Particularly, temporary storage 6 is a noise information with the noise spectrum shape.Yet temporary storage 6 for example can also replace frequency spectrum perhaps to store with frequency spectrum the frequency characteristic of phase place and the characteristic storage such as intensity and time variation of CF.Noise information can also comprise statistical information (maximum, minimum, variance and intermediate value) etc.
Noise suppression unit 3 is utilized the deterioration signal amplitude spectrum that provided by FFT unit 2 and is suppressed the noise at each frequency place by the noise information that temporary storage 6 provides, and the signal amplitude spectrum that enhancing is provided to IFFT unit 4 is as the squelch result.Inverse transformation is carried out in the combination of enhancing signal amplitude spectrum that 4 pairs of IFFT unit provide from noise suppression unit 3 and the deterioration signal phase place that provides from FFT unit 2, and the enhancing signal sample is offered lead-out terminal 5.
Noise information generation unit 7 also is provided as this enhancing signal amplitude spectrum that suppresses the result simultaneously.Noise information generation unit 7 generates new noise information based on this enhancing signal amplitude spectrum as the squelch result, and new noise information is offered temporary storage 6.Temporary storage 6 utilizes the new noise information that provides from noise information generation unit 7 to upgrade current noise information.
< layout of FFT unit 2 >
Fig. 2 is the block diagram that the layout of FFT unit 2 is shown.As shown in Figure 2, FFT unit 2 comprises frame division unit 21, adds window unit 22 and fourier transform unit 23.Frame division unit 21 receives the deterioration signal sample and it is divided into the frame corresponding to K/2 sample, and wherein K is an even number.The deterioration signal sample that is divided into frame is provided for and adds window unit 22 and multiply by window function w (t).Through w (t) give n frame input signal yn (t) (t=0,1 ..., K/2-1) windowing and the signal that obtains is provided by following formula
y &OverBar; n ( t ) = w ( t ) y n ( t ) . . . ( 1 )
What extensively carried out equally is to two successive frame windowings of part overlaid (overlapping) each other.Suppose that overlapping length is 50% of frame length.For t=0; 1; ...; K/2-1 adds and
Figure BPA00001547322600053
that window unit 22 outputs are provided by following formula
y &OverBar; n ( t ) = w ( t ) y n - 1 ( t + K / 2 ) y &OverBar; n ( t + K / 2 ) = w ( t + K / 2 ) y n ( t ) . . . ( 2 )
The symmetry-windows function is used to real signal.When spectrum gain in MMSE STSA method is set as 1 or in the SS method zero when being deducted, except error, window function makes input signal and output Signal Matching.This means w (t)=w (t+K/2)=1.
To continue to describe example below to 50% two the successive frame windowings of overlapping.Add window unit 22 and for example can use peaceful (hanning) window w (t) of the Chinese that provides by following formula
Figure BPA00001547322600055
Alternately, add window unit 22 and can use various window functions, for example Hamming (hamming) window, Caesar's (Kaiser) window and Blacknam (Blackman) window.Output after the windowing is provided for fourier transform unit 23 and is transformed to deterioration signal spectrum Yn (k).Deterioration signal spectrum Yn (k) is divided into phase place and amplitude.Deterioration signal phase spectrum arg Yn (k) is provided for IFFT unit 4, and the deterioration signal amplitude spectrum | Yn (k) | be provided for noise suppression unit 3.As described, FFT unit 2 can use power spectrum to replace amplitude spectrum.
< layout of IFFT unit 4 >
Fig. 3 is the block diagram that the layout of IFFT unit 4 is shown.As shown in Figure 3, IFFT unit 4 comprises inverse Fourier transform unit 43, adds window unit 42 and frame reconfiguration unit 41.Inverse Fourier transform unit 43 will be from the combined enhancing signal to obtain to be provided by following formula of noise suppression unit 3 enhancing signal amplitude spectrum that provides and the deterioration signal phase spectrum arg Yn (k) that provides from FFT unit 2
X &OverBar; n ( k ) = | X &OverBar; n ( k ) | &CenterDot; arg Y n ( k ) . . . ( 4 )
The 43 pairs of enhancing signal that obtain in inverse Fourier transform unit are carried out inverse Fourier transform.Enhancing signal after the inverse Fourier transform be provided for add window unit 42 as time domain samples sequence xn (t) (t=0,1 ..., K-1), wherein, a frame comprises K sample and multiply by window function w (t).Through with w (t) to n frame input signal xn (t) (t=0,1 ..., K/2-1) windowing and the signal that obtains are provided by following formula
x &OverBar; n ( t ) = w ( t ) x n ( t ) . . . ( 5 )
What extensively carried out equally is to two successive frame windowings of part overlaid (overlapping) each other.Suppose that overlapping length is 50% of frame length.For t=0; 1; ...; K/2-1 adds
Figure BPA00001547322600063
and
Figure BPA00001547322600064
that window unit 42 outputs are provided by following formula
x &OverBar; n ( t ) = w ( t ) x n - 1 ( t + K / 2 ) x &OverBar; n ( t + K / 2 ) = w ( t + K / 2 ) x n ( t ) . . . ( 6 )
And they are provided to frame reconfiguration unit 41.
Frame reconfiguration unit 41 is directed against every K/2 sample from adding the output that window unit 42 extracts two consecutive frames; Overlapping they; And for t=0,1 ...; K-1 obtains the output signal
Figure BPA00001547322600066
that is provided by following formula
x ^ n ( t ) = x &OverBar; n - 1 ( t + K / 2 ) + x &OverBar; n ( t ) . . . ( 7 )
Frame reconfiguration unit 41 provides the output that obtains signal to lead-out terminal 5.
Notice that the FFT unit 2 among Fig. 2 and Fig. 3 is described to Fourier transform in the above with the conversion in the IFFT unit 4.Yet FFT unit 2 can use any other conversion with IFFT unit 4, and for example cosine transform, warp are revised discrete cosine transform (MDCT), hadamard (Hadamard) conversion, Ha Er (Haar) conversion or wavelet transformation, replace Fourier transform.For example, cosine transform or warp are revised cosine transform and are only obtained amplitude as transformation results.This got rid of among Fig. 1 from the needs in the path of 2 to IFFT unit 4, FFT unit.In addition, the noise information that is recorded in the temporary storage 6 only need comprise amplitude (or power), thereby helps to reduce the calculation times of memory size and squelch processing.Haar transform allows contraction in multiplication and reduces the LSI area of chip.Because wavelet transformation can depend on the frequency shift temporal resolution, therefore can expect better squelch.
Alternately, after a plurality of frequency contents had been integrated in FFT unit 2, noise suppression unit 3 can be carried out actual inhibition.In this situation, FFT unit 2 can multi-frequency becomes to assign to obtain the high sound quality through integrate more from low-frequency range, wherein, hears in the low-frequency range that the discrimination of characteristic is higher than the high-frequency range with relatively poor ability.When after having integrated a plurality of frequency contents, carrying out squelch, be applied in the frequency content decreased number of squelch.Therefore Noise Suppression Device 100 can reduce the overall calculation number of times.
< processing of noise suppression unit 3 >
Noise suppression unit 3 can be carried out various inhibition.Typical inhibition method is SS (spectrum-subtraction) method and MMSE STSA (least mean-square error short-time spectrum amplitude estimation device) method.When using the SS method, noise suppression unit 3 deducts the noise information that is provided by temporary storage 6 from the deterioration signal amplitude spectrum that is provided by FFT unit 2.When using MMSE STSA method, noise suppression unit 3 is utilized the noise information that provided by temporary storage 6 and is calculated rejection coefficient to each frequency content in a plurality of frequency contents by the deterioration signal amplitude spectrum that FFT unit 2 provides.Noise suppression unit 3 multiply by this rejection coefficient with the deterioration signal amplitude spectrum.This rejection coefficient is confirmed as and makes the mean square power of enhancing signal minimize.
Noise suppression unit 3 can be used limit value (flooring) and avoid excessive noise to suppress.Limit value is a kind of method of avoiding suppressing to surpass the largest inhibition amount.The limit value parameter is confirmed the largest inhibition amount.When using the SS method, noise suppression unit 3 applies restriction, therefore is not less than the limit value parameter through the result who deducts from the deterioration signal amplitude spectrum through revising the noise information acquisition.More specifically, if subtraction result less than the limit value parameter, then noise suppression unit 3 utilizes the limit value parameter to replace subtraction result.When using MMSE STSA method, if from gaining less than the limit value parameter through the spectrum of revising the acquisition of noise information and deterioration signal amplitude spectrum, then noise suppression unit 3 utilizes the limit value parameter to replace the spectrum gain.The details of limit value is disclosed in document " M.Berouti, R.Schwartz, and J.Makhoul; " Enhancement of speech corrupted by acoustic noise "; Proceedings of ICASSP ' 79, pp.208-211, Apr.1979 ".When limit value was introduced into, noise suppression unit 3 can not carried out excessive inhibition.Limit value can prevent that enhancing signal has big distortion.
Noise suppression unit 3 can also be provided with the frequency content number of noise information less than the frequency content number of deterioration signal spectrum.At this moment, a plurality of frequency contents are shared many noise informations.Compare with a plurality of frequency contents of situation integrate to(for) deterioration signal spectrum and noise information, the frequency resolution of deterioration signal spectrum is higher.Therefore, noise suppression unit 3 can be through obtaining high sound quality than the calculated amount of lacking in the situation that does not have frequency content to integrate.Jap.P. in early days openly No.2008-203879 the details that the noise information that uses its frequency content number to lack than the frequency content number of deterioration signal spectrum suppresses is disclosed.
< layout of noise information generation unit 7 >
Enhancing signal amplitude spectrum as the squelch result is provided for noise information generation unit 7.Noise information generation unit 7 utilizes the squelch result to generate new noise information, and utilizes new noise information to update stored in the noise information in the temporary storage 6.For example, the even shape signal spectrum is prepared as the default value that is stored in the noise information in the temporary storage 6.Noise information generation unit 7 depends on that the squelch result that wherein said signal spectrum is used as noise information generates new noise information.Noise information generation unit 7 updates stored in the noise information that being used in the temporary storage 6 suppress.
When the squelch result who feeds back to noise information generation unit 7 when utilization generates new noise information; Noise information generation unit 7 generted noise information are not so that there is the squelch result of timing place of echo signal input bigger (do not have repressed residual noise big more), and noise information is just big more.There is not the big squelch result of timing place of echo signal input to indicate inhibition not enough.Therefore, preferably make noise information bigger.When noise information was big, the subtrahend value of SS method was bigger, and therefore the squelch result diminishes.In the inhibition such as MMSE STSA method based on multiplication, the signal to noise ratio (snr) that is used to calculate rejection coefficient is estimated less, and therefore, little rejection coefficient can be obtained.This has caused stronger squelch.Several different methods can be used to generate new noise information.With describe as an example recomputate algorithm and recurrence update algorithm.
In desirable squelch result, noise is suppressed fully.Noise information generation unit 7 for example can hour recomputate or recursively upgrades noise information at the amplitude of deterioration signal or power, so that suppress noise fully.This is higher because leave no choice but the little probability of the power of repressed noise when the amplitude of deterioration signal or power hour signal.Can utilize noise information generation unit 7 absolute value or the little fact of power ratio threshold value of the amplitude of deterioration signal detect deterioration signal by a small margin or power.
Noise information generation unit 7 can also utilize the noise information that is recorded in the temporary storage 6 and deterioration signal amplitude or power difference less than the fact of threshold value detect deterioration signal by a small margin or power.That is, noise information generation unit 7 utilizes following true: when the amplitude of deterioration signal or power were similar to noise information, noise information was formed most (SNR is lower) of deterioration signal.Especially, noise information generation unit 7 can utilize the combination of the information at a plurality of Frequency points place to come the comparison spectrum envelope, detects degree of accuracy thereby improve.
Noise information in the SS method is recalculated as being each frequency for timing place that does not have the echo signal input, equals the deterioration signal amplitude spectrum.In other words, noise information generation unit 7 makes the deterioration signal amplitude spectrum that when only noise is transfused to, provides from FFT unit 2 | Yn (k) | be complementary with noise information ν n (k).That is, noise information generation unit 7 comes calculating noise information ν n (k) through utilizing following formula
νn(k)＝|Yn(k)| ...(8)
Wherein, n is a frame number, and k is a frequency number.
Noise information generation unit 7 can use average the substituting of noise information ν n (k) directly to use noise information ν n (k).This on average can be based on average (the using the moving average of sliding window) of FIR wave filter or based on average (the leakage integration) of iir filter.
On the other hand, the noise information recurrence in the SS method upgrade to be through upgrading noise information gradually to realize so that for each frequency, approach zero at the enhancing signal amplitude spectrum of timing place that does not have the echo signal input.When perturbation motion method being used for the recurrence renewal, noise information generation unit 7 utilizes the error e n (k) of n frame to calculate ν n+1 (k) to frequency number k, as follows
νn+1(k)＝νn(k)+μen(k) ...(9)
Wherein, μ is the small constant (microconstant) that is called step size.If will be used immediately through calculating the noise information ν n (k) that obtains, then noise information generation unit 7 uses
νn(k)＝νn-1(k)+μen(k)
...(10)
Replace equality (9).That is, noise information generation unit 7 utilizes error current to calculate current noise information ν n (k) and uses it immediately.Noise information generation unit 7 can be realized real-time accurate squelch through upgrading noise information immediately.
Alternately, noise information generation unit 7 can be as follows like this utilization only represent the signum sgn{en (k) of the symbol of error come calculating noise information ν n+1 (k)
νn+1(k)＝νn(k)+μ·sgn{en(k)}
...(13)
Similarly, noise information generation unit 7 can use any other update algorithm (recurrence update algorithm).
When using MMSE STSA method, noise information generation unit 7 recursively upgrades noise information.Noise information generation unit 7 is through upgrading noise information ν n (k) to each frequency with utilizing the described same procedure of equality (9) to (11).
Property feature as recomputating as noise information update method above-mentioned with the recurrence update algorithm recomputates algorithm and have the high speed of following, and the recurrence update algorithm has pinpoint accuracy.In order to utilize these property features, noise information generation unit 7 can change update method so that for example at first use and recomputate algorithm, uses the recurrence update algorithm then.When confirming to change the timing of update method, noise information generation unit 7 can change update method under the situation near optimal value fully at noise information.Change update method when alternately, noise information generation unit 7 for example can be over and done with at the fixed time.Noise information generation unit 7 changes update method when in addition, can drop under the predetermined threshold in the index word of noise information.
As stated, the Noise Suppression Device 100 of this exemplary embodiment generates the noise information that is used for squelch based on the squelch result.Therefore, the various noises that suppress to comprise unknown noise under the situation of many noise informations can need not to store in advance.
(second exemplary embodiment)
Second exemplary embodiment will be described.The noise information generation unit 7 of second exemplary embodiment multiply by zoom factor and generates noise information through will for good and all being stored in essential information in the nonvolatile memory.For example, any information as the even shape signal spectrum is prepared the essential information (default value) as noise information.Noise information generation unit 7 generates noise information through essential information multiply by zoom factor, and after this, upgrades noise information and zoom factor thereof according to the squelch result who uses noise information.Being updated in first exemplary embodiment of noise information described in detail.Therefore the renewal of zoom factor is described here.
When utilizing the squelch result to generate zoom factor, noise information generation unit 7 generates zoom factors so that there is not the squelch result of timing place of echo signal input bigger (do not have repressed residual noise big more), and noise information is just big more.There is not the big squelch result of timing place of echo signal input to indicate inhibition not enough.Therefore, preferably make noise information bigger through changing zoom factor.Several different methods can be used to upgrade zoom factor.With describe as an example recomputate algorithm and recurrence update algorithm.
In desirable squelch result, noise is suppressed fully.Noise information generation unit 7 for example can hour recomputate or recursively upgrades zoom factor at the amplitude of deterioration signal or power, so that suppress noise fully.This is higher because leave no choice but the little probability of the power of repressed noise when the amplitude of deterioration signal or power hour signal.Can utilize noise information generation unit 7 absolute value or the little fact of power ratio threshold value of the amplitude of deterioration signal detect deterioration signal by a small margin or power.
Noise information generation unit 7 can also utilize the noise information that is recorded in the temporary storage 6 and deterioration signal amplitude or power difference less than the fact of threshold value detect deterioration signal by a small margin or power.That is, noise information generation unit 7 utilizes following true: when the amplitude of deterioration signal or power were similar to noise information, noise was formed most (SNR is lower) of deterioration signal.Especially, noise information generation unit 7 can utilize the combination of the information at a plurality of Frequency points place to come the comparison spectrum envelope, detects degree of accuracy thereby improve.
Zoom factor in the SS method is recomputated so that for each frequency of timing place that does not have echo signal input, noise information equals the deterioration signal amplitude spectrum.In other words, noise information generation unit 7 obtains convergent-divergent mark α n (k) so that the deterioration signal amplitude spectrum that when only noise is transfused to, provides from FFT unit 2 | Yn (k) | be complementary with zoom factor α n and the amassing of essential information ν n (k).That is, zoom factor α n (k) passes through computes
αn(k)＝|Yn(k)|/ν(k) ...(12)
Wherein, n is a frame number, and k is a frequency number.
On the other hand, the zoom factor recurrence in the SS method upgrade to be through upgrading zoom factor gradually to realize so that for each frequency, approach zero at the enhancing signal amplitude spectrum of timing place that does not have the echo signal input.When LMS (least squares approach) algorithm being used for the recurrence renewal, noise information generation unit 7 utilizes the error e n (k) of n frame to come calculation of alpha n+1 (k) to frequency number k, as follows
αn+1(k)＝αn(k)+μen(k)ν(k) ...(13)
Wherein, μ is the small constant (microconstant) that is called step size.If will be used by Noise Suppression Device 100 immediately through calculating the zoom factor α n (k) that obtains, then noise information generation unit 7 uses
αn(k)＝αn-1(k)+μen(k)ν(k) ...(14)
Replace equality (13).That is, noise information generation unit 7 utilizes error current to calculate current zoom factor α n (k) and is applied to Noise Suppression Device 100 immediately.Noise information generation unit 7 can be realized real-time accurate squelch through upgrading zoom factor immediately.
When using NLMS (normalization least squares approach) algorithm, noise information generation unit 7 utilizes above-mentioned error e n (k) to calculate zoom factor α n+1 (k) as follows like this
αn+1(k)＝αn(k)+μen(k)ν(k)/σn(k) 2 ...(15)
Wherein, σ n (k) 2Be the average power of noise information ν n (k), it can utilize based on average (moving average of use sliding window) of FIR wave filter or based on average (the leakage integration) of iir filter and calculated.
Noise information generation unit 7 can utilize method of perturbation to calculate zoom factor α n+1 (k) as follows like this
αn+1(k)＝αn(k)+μen(k) ...(16)
Alternately, noise information generation unit 7 can be as follows like this utilization only represent the signum sgn{en (k) of the symbol of error calculate zoom factor α n+1 (k)
αn+1(k)αn(k)+μ·sgn{en(k)} ...(17)
Similarly, noise information generation unit 7 can use LS (least square) algorithm or any other update algorithm.Noise information generation unit 7 can also be used the zoom factor through upgrading immediately.In this situation, the implementor of Noise Suppression Device 100 can revise equality (15) to (17) from equality (13) to the change of equality (14) through reference thereby noise information generation unit 7 is designed to upgrade in real time zoom factor.
Use MMSE STSA method, noise information generation unit 7 recursively upgrades zoom factor.Noise information generation unit 7 is through upgrading zoom factor α n (k) to each frequency with utilizing the described same procedure of equality (13) to (17).
Property feature as recomputating as zoom factor update method above-mentioned with the recurrence update algorithm recomputates algorithm and have the high speed of following, and the recurrence update algorithm has pinpoint accuracy.In order to utilize these property features, noise information generation unit 7 can change update method so that for example at first use and recomputate algorithm, uses the recurrence update algorithm then.Noise information generation unit 7 can change update method under the situation near optimal value fully at zoom factor.Change update method when alternately, noise information generation unit 7 for example can be over and done with at the fixed time.Noise information generation unit 7 changes update method when in addition, can drop under the predetermined threshold in the index word of zoom factor.
In this exemplary embodiment, identical in the layout the noise information generation method in noise information generation unit 7 and operation and first exemplary embodiment so will do not repeated the description to it.
May consider, noise information be important information and in the renewal of noise information and scalability information scalability information will be modified.Noise information generation unit 7 can upgrade noise information and upgrade scalability information to little change to big change.Particularly, from the information processing of default value generted noise, through upgrading noise information generted noise information fast.When noise information reduced near right value and error, the accurate output of noise information generation unit can be obtained through upgrading scalability information.
According to this exemplary embodiment; Except the effect of first exemplary embodiment, can also follow the change of noisiness apace and obtain the accurate output of noise information generation unit through the renewal and more the newly arriving of scalability information of making up noise information alternatively.
(the 3rd exemplary embodiment)
To the 3rd exemplary embodiment be described with reference to figure 4.Except the layout of first exemplary embodiment, Noise Suppression Device 200 also comprises input terminal 9.Noise suppression unit 53 receives the expression specific noise with noise information generation unit 47 from input terminal 9 and whether is present in the information (noise existence information) the deterioration signal of being imported.Thus, Noise Suppression Device 200 can make it possible to suppress noise and while generted noise information reliably in timing place that specific noise exists.Identical in all the other layouts and operation and first exemplary embodiment, so will not repeat detailed description to it.
The Noise Suppression Device 200 of this exemplary embodiment generted noise information not when specific noise does not exist.Therefore, can obtain higher squelch degree of accuracy to specific noise.
(the 4th exemplary embodiment)
To the 4th exemplary embodiment be described with reference to figure 5.The Noise Suppression Device 300 of this exemplary embodiment comprises echo signal detecting unit 51.FFT unit 2 provides the deterioration signal amplitude spectrum to echo signal detecting unit 51.Echo signal detecting unit 51 judges whether echo signal exists or exist degree in this deterioration signal amplitude spectrum.
The result of determination of based target detecting signal unit 51, noise information generation unit 57 generted noise information.For example, when not having echo signal, deterioration signal only comprises noise, so the inhibition result of noise suppression unit 3 is necessary for zero.Therefore, noise information generation unit 57 is adjusted in noise information of describing in first exemplary embodiment and the zoom factor of in second exemplary embodiment, describing, to obtain the zero squelch result as this moment.
On the other hand, when deterioration signal comprised echo signal, noise information generation unit 57 generated noise information according to the ratio that exists of echo signal.For example, be 10% if be present in the ratio of the echo signal in the deterioration signal, then noise information generation unit 57 partly (only 90%) update stored in the noise information in the temporary storage 6.
The Noise Suppression Device 300 of this exemplary embodiment generates noise information according to the ratio of the noise in the deterioration signal.This allows to obtain more accurate squelch result.
(the 5th exemplary embodiment)
To the 5th exemplary embodiment be described with reference to figure 6.Fig. 6 is the block diagram that the signal conditioning package 500 that is included in the Noise Suppression Device of describing in first exemplary embodiment 400 is shown.Signal conditioning package 500 comprises the mechanical control unit 92 as the machine assembly 91 of noise source and control machine assembly 91.Former thereby when operating machine unit 91 because of certain when mechanical control unit 92, Noise Suppression Device 400 is provided operation information.This allows Noise Suppression Device 400 to operate reliably in the operating period of machine assembly 91 and generates noise information.
Alternately; Mechanical control unit 92 can be based on the unit 91 of operating machine from the instruction of the generted noise of Noise Suppression Device 400; And simultaneously, the noise information generation unit 67 in the Noise Suppression Device 400 can utilize the deterioration signal that comprises noise to generate noise information.
(other exemplary embodiment)
About Noise Suppression Device first to the 5th exemplary embodiment has been described above with different qualities characteristic.Exemplary embodiment also comprises the Noise Suppression Device that forms through with mode combined characteristic characteristic whatsoever.
The present invention can be applicable to comprise the system or the single assembly of a plurality of equipment.When the signal handler of the software of the function that is used for realization example property embodiment directly or when remote site was loaded into system or device, the present invention was applicable equally.Therefore, the present invention also comprises and being installed in the computing machine so that computer realization functional programs of the present invention, store the medium of this program and download the Web server of this program from it.
Fig. 7 is a block diagram of carrying out the computing machine 1000 of the signal handler that is configured to first to the 5th exemplary embodiment.Computing machine 1000 comprises input block 1001, CPU 1002, output unit 1003, storer 1004, external memory storage 1005, communication control unit 1006 and the bus that is connected these 1007.
CPU 1002 comes the operation of control computer 1000 through the read output signal handling procedure.More specifically, when carrying out signal handler, CPU 1002 suppresses the noise in the deterioration signals and generates noise information (S801) based on the squelch result.Next, CPU 1002 utilizes institute's generted noise information to suppress the noise (S802) in the deterioration signal.Do not generated (S804) as yet if action stops incident, then CPU 1002 utilizes the squelch result to upgrade noise information (S803).That is, CPU 1002 repeatedly carries out noise information generation/renewal and squelch till action stops incident and is transfused to.Comprise that the outage and the exercises of mute microphone (MIC) stop incident and taked.
Aforesaid computing machine make it possible to obtain with first to the 5th exemplary embodiment in identical effect.
Though reference example property embodiment has described the present invention above, yet the invention is not restricted to exemplary embodiment.It will be apparent to one skilled in the art that and under the situation that does not break away from the spirit and scope of the present invention, to revise layout of the present invention and details in every way.
The application is based on the Japanese patent application No.2009-255419 that submitted on November 6th, 2009 and require its right of priority, and the open of this application is incorporated into this by integral body by reference.

Claims (8)

1. signal processing method that is used for suppressing the noise of deterioration signal comprises:
Squelch result according to said deterioration signal generates noise information; And
Utilize the noise information that is generated to suppress the noise in the said deterioration signal.
2. signal processing method according to claim 1, wherein, said noise information generates through essential information multiply by zoom factor.
3. signal processing method according to claim 1 and 2, wherein, the information whether the expression noise is present in the said deterioration signal is transfused to, and said noise information is generated when noise is present in the said deterioration signal.
4. according to each described signal processing method in the claim 1 to 3, wherein, through analyze said deterioration signal confirm in the said deterioration signal echo signal have a degree, and generate said noise information based on definite result.
5. signal conditioning package comprises:
Noise suppressor, this noise suppressor suppresses the noise in the deterioration signal; And
Noise information generation unit, this noise information generation unit be based on generating noise information to the Noise Suppression result in the said deterioration signal,
Wherein, said noise suppressor utilizes said noise information to suppress the noise in the said deterioration signal.
6. signal conditioning package according to claim 5 also comprises: storage unit, the said noise information that this cell stores is generated by said noise information generation unit.
7. according to claim 5 or 6 described signal conditioning packages, also comprise:
Machine assembly, said machine assembly is as noise source; And
Mechanical control unit, said mechanical control unit is controlled said machine assembly,
Wherein, in timing place of said mechanical control unit through the said machine assembly generted noise of operation, said noise information generation unit generates said noise information.
8. computer-readable medium that is used for the storage signal handling procedure, said signal handler make computing machine carry out following the processing:
Generate noise information and noise information that utilization generated suppresses the noise in the deterioration signal based on the process result that suppresses noise.
CN201080050449.9A 2009-11-06 2010-11-02 Signal processing method, information processor Active CN102598127B (en)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
JP2009-255419 2009-11-06
JP2009255419A JP2011100029A (en) 2009-11-06 2009-11-06 Signal processing method, information processor, and signal processing program
PCT/JP2010/069869 WO2011055829A1 (en) 2009-11-06 2010-11-02 Signal processing method, information processor, and signal processing program

Publications (2)

Publication Number Publication Date
CN102598127A true CN102598127A (en) 2012-07-18
CN102598127B CN102598127B (en) 2016-07-13

Family

ID=43970061

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201080050449.9A Active CN102598127B (en) 2009-11-06 2010-11-02 Signal processing method, information processor

Country Status (5)

Country Link
US (1) US9190070B2 (en)
EP (1) EP2498251B1 (en)
JP (1) JP2011100029A (en)
CN (1) CN102598127B (en)
WO (1) WO2011055829A1 (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106910511A (en) * 2016-06-28 2017-06-30 阿里巴巴集团控股有限公司 A kind of speech de-noising method and apparatus
CN107045872A (en) * 2016-02-05 2017-08-15 中国电信股份有限公司 The recognition methods of talk echo and device

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR3034625A1 (en) 2015-04-10 2016-10-14 Naturex EUTECTIC EXTRACTION SOLVENT, EUTECTIGENESE EXTRACTION METHOD USING THE SOLVENT, AND EXTRACT FROM THE EXTRACTION PROCESS.

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4630304A (en) * 1985-07-01 1986-12-16 Motorola, Inc. Automatic background noise estimator for a noise suppression system
EP1065656A2 (en) * 1994-05-13 2001-01-03 Sony Corporation Method for reducing noise in an input speech signal
CN1741133A (en) * 2004-08-27 2006-03-01 日本电气株式会社 Sound processing apparatus and input sound processing method
WO2007058121A1 (en) * 2005-11-15 2007-05-24 Nec Corporation Reverberation suppressing method, device, and reverberation suppressing program
CN101111888A (en) * 2005-02-02 2008-01-23 富士通株式会社 Signal processing method and device thereof

Family Cites Families (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH05316186A (en) * 1992-05-08 1993-11-26 Matsushita Electric Ind Co Ltd Voice recognition telephone set
ZA95599B (en) * 1994-02-28 1996-02-06 Qualcomm Inc Doubletalk detection by means of spectral content
JP3451146B2 (en) 1995-02-17 2003-09-29 株式会社日立製作所 Denoising system and method using spectral subtraction
US6001131A (en) * 1995-02-24 1999-12-14 Nynex Science & Technology, Inc. Automatic target noise cancellation for speech enhancement
JPH09212196A (en) * 1996-01-31 1997-08-15 Nippon Telegr & Teleph Corp <Ntt> Noise suppressor
US6570985B1 (en) * 1998-01-09 2003-05-27 Ericsson Inc. Echo canceler adaptive filter optimization
US6850783B1 (en) * 1998-08-07 2005-02-01 Ericsson Inc. Methods and apparatus for mitigating the effects of microphone overload in echo cancelation systems
US6606382B2 (en) * 2000-01-27 2003-08-12 Qualcomm Incorporated System and method for implementation of an echo canceller
JP2001215990A (en) * 2000-01-31 2001-08-10 Japan Science & Technology Corp Robot hearing device
US6859531B1 (en) * 2000-09-15 2005-02-22 Intel Corporation Residual echo estimation for echo cancellation
JP4282227B2 (en) 2000-12-28 2009-06-17 日本電気株式会社 Noise removal method and apparatus
US7349841B2 (en) * 2001-03-28 2008-03-25 Mitsubishi Denki Kabushiki Kaisha Noise suppression device including subband-based signal-to-noise ratio
JP4088148B2 (en) * 2002-12-27 2008-05-21 松下電器産業株式会社 Noise suppressor
US6947549B2 (en) * 2003-02-19 2005-09-20 The Hong Kong Polytechnic University Echo canceller
DE602005026426D1 (en) * 2004-11-02 2011-03-31 Panasonic Corp NOISE IN HOUR
JP4697465B2 (en) * 2004-11-08 2011-06-08 日本電気株式会社 Signal processing method, signal processing apparatus, and signal processing program
JP2006279185A (en) 2005-03-28 2006-10-12 Casio Comput Co Ltd Imaging apparatus, and sound recording method and program
JP4639907B2 (en) * 2005-03-31 2011-02-23 カシオ計算機株式会社 Imaging apparatus, audio recording method, and program
US9318119B2 (en) 2005-09-02 2016-04-19 Nec Corporation Noise suppression using integrated frequency-domain signals
JP2007116585A (en) * 2005-10-24 2007-05-10 Matsushita Electric Ind Co Ltd Noise cancel device and noise cancel method
US20080059154A1 (en) * 2006-09-01 2008-03-06 Nokia Corporation Encoding an audio signal
JP2009255419A (en) 2008-04-17 2009-11-05 Toppan Cosmo Inc Decorative sheet
JP2009276528A (en) * 2008-05-14 2009-11-26 Yamaha Corp Sound processor and recording device

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4630304A (en) * 1985-07-01 1986-12-16 Motorola, Inc. Automatic background noise estimator for a noise suppression system
EP1065656A2 (en) * 1994-05-13 2001-01-03 Sony Corporation Method for reducing noise in an input speech signal
CN1741133A (en) * 2004-08-27 2006-03-01 日本电气株式会社 Sound processing apparatus and input sound processing method
CN101111888A (en) * 2005-02-02 2008-01-23 富士通株式会社 Signal processing method and device thereof
WO2007058121A1 (en) * 2005-11-15 2007-05-24 Nec Corporation Reverberation suppressing method, device, and reverberation suppressing program

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107045872A (en) * 2016-02-05 2017-08-15 中国电信股份有限公司 The recognition methods of talk echo and device
CN107045872B (en) * 2016-02-05 2020-09-01 中国电信股份有限公司 Recognition method and device of call echo
CN106910511A (en) * 2016-06-28 2017-06-30 阿里巴巴集团控股有限公司 A kind of speech de-noising method and apparatus
CN106910511B (en) * 2016-06-28 2020-08-14 阿里巴巴集团控股有限公司 Voice denoising method and device

Also Published As

Publication number Publication date
US20120207326A1 (en) 2012-08-16
WO2011055829A1 (en) 2011-05-12
CN102598127B (en) 2016-07-13
JP2011100029A (en) 2011-05-19
EP2498251B1 (en) 2019-12-25
US9190070B2 (en) 2015-11-17
EP2498251A4 (en) 2013-08-07
EP2498251A1 (en) 2012-09-12

Similar Documents

Publication Publication Date Title
US9431023B2 (en) Monaural noise suppression based on computational auditory scene analysis
US10123113B2 (en) Selective audio source enhancement
CN103109320B (en) Noise suppression device
US9837097B2 (en) Single processing method, information processing apparatus and signal processing program
CN102612711B (en) Signal processing method, information processor
EP3329488B1 (en) Keystroke noise canceling
WO2012158156A1 (en) Noise supression method and apparatus using multiple feature modeling for speech/noise likelihood
TW201248613A (en) System and method for monaural audio processing based preserving speech information
EP2752848A1 (en) Method and apparatus for generating a noise reduced audio signal using a microphone array
Martín-Doñas et al. Dual-channel DNN-based speech enhancement for smartphones
CN102598126A (en) Information processing device, auxiliary device therefor, information processing system, control method therefor, and control program
CN102598127A (en) Signal processing method, information processor, and signal processing program
US20140249809A1 (en) Audio signal noise attenuation
CN113035216B (en) Microphone array voice enhancement method and related equipment
Diaz‐Ramirez et al. Robust speech processing using local adaptive non‐linear filtering
CN102598128A (en) Signal processing method, information processor, and signal processing program
CN103270772B (en) Signal handling equipment, signal processing method
CN109378012B (en) Noise reduction method and system for recording audio by single-channel voice equipment
US11176957B2 (en) Low complexity detection of voiced speech and pitch estimation
CN116504264B (en) Audio processing method, device, equipment and storage medium
Dionelis On single-channel speech enhancement and on non-linear modulation-domain Kalman filtering
Esch et al. Model-based speech enhancement exploiting temporal and spectral dependencies
Seyedin et al. Robust MVDR-based feature extraction for speech recognition
Abd Almisreb et al. Noise reduction approach for Arabic phonemes articulated by Malay speakers
CN114360572A (en) Voice denoising method and device, electronic equipment and storage medium

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant